A widespread myth of sensor resolution (or the end of the megapixel race) states that if you print larger, you will step back further and, consequently, will never need the resolution given by higher megapixel cameras and prints. This myopic view of megapixels is correct if you are thinking of snapshots, looking solely at web uploads, or preventing* people from getting close enough to inspect the images.
* Helmut Newton’s big nudes are hung above the stairs in the lobby of the Berlin Museum, where it is physically impossible to get closer than about 2 meters, to realize that they are really not that sharp.
The images in my web gallery are heavily compressed jpegs of about 5 MP. An Apple Thunderbolt Display runs at 109 ppi (2560 x 1440) while the highest resolution display is the one of the iPad Mini retina at 326 ppi (2048 x 1536). My images therefore qualify to be viewed at full size on the iPad and the Thunderbolt displays. As long as you can still focus that close, you can stick your nose to the iPad’s screen without beeing able to distinguish single pixels. But the images look less than optimal on the Thunderbolt when viewed from less than about a foot (30 cm) away. There is simply no way a screen can do an image justice. With no universal standard for monitor calibration, it is difficult to ensure that you will see on your monitor what I did on mine.
On the same footing that I warn my friends about photographing their kids exclusively using the iPhone, I should review my own work. For more than 20 years I had my work printed on Cibachrome, because there was simply no other way of displaying the large-format images, if we forget the loupe and light table. And pre-viewing b&w and color reversal film was impossible altogether. I have to admit, however, that pixel-peeping has more and more replaced this habit after the advent of film scanning and the all-digital workflow. But some recent visits to galleries made me think: I saw scanned film made to look like collodion wet-plate, film made to look digital, digital made to look like Cibachrome, and sub-inch sensor digital blown up to gallery-size. And all that with limited success; mediocre printing techniques as an artistic statement, a sign of carelessness, or simply bad craftsmanship?
Making a print is the final part of the photography workflow. Without a print, there is no reference on how an image should be viewed and what quality of capture is required. I would define a high-impact, technically perfect print as evoking the following reaction among the viewers: they stop at a distance where they can see the whole image, say 2 meters for a 29” (740 mm) print. Realizing the image is sharp and detailed, they move in closer and closer as long as they can still focus. This is what I call supernatural, because the print reveals details in extend of the visual acuity at the location. So the question is: what is the point of sufficiency, or how much is enough?
Visual acuity, a measure for distinguishing details, is defined as the reciprocal of the angular distance that must separate two contours so that they appear as discrete; a = 1/r, where a is the (dimensionless) acuity and r the response, expressed arc-minutes per line pair. The acuity is 1.1-1.7 under good lighting conditions*. The acuity of 1.5 corresponds to 0.66 arc minutes per line pair. For a minimum of two pixels per line pair, we have a pixel spacing of 0.33 arc-minutes, that is, 0.33*Pi/10800 radians = 0.1 mrad or 0.1 mm (an eyelash) at one meter distance.
*S. Hecht: The Relation between Visual Acuity and Illumination, The Journal of General Physiology, 1927
Consider now viewing an A1 print (590 x 740 mm, 23 x 29″) from 29” distance. The wide side of the print subtends an angle of 0.9 radian, or 53 deg, which roughly corresponds to the picture angle of my 45 mm PC-E lens. At least 10000 pixels are therefore required to show detail at the limits of the human visual acuity. This corresponds to a printer resolution of 345 ppi, which is roughly that of the iPad Mini retina, albeit at a 37” diagonal instead of the 7.9” of the iPad Mini.
For the A1 print, a minimum of 80 MP is therefore required, the Nikon D800E will do for an A2 print, and an APS-C sized sensor for an A3. So size does matter. However, a 64” inch (1.6 m) print viewed at 64” distance still requires 10000 pixels on its long side. This is good, because we do not need more resolution as long as the viewing angle remains constant. But if we choose to examine only a small portion of this print we would be disappointed to see pixelation, print dots, and smudging. And this also means that on subjects with semi-fractal detail, such as forests and desert landscapes, we can see a difference between 16 and 36 MP even at print sizes as small as A4.
That’s just pure theory, one might say. So let us examine what all this means in practice. For the sake of scientific reasoning and for not wasting too much ink and paper, here is a proposal of a test sequence, up for discussion:
- Check how print quality can be discussed and judged objectively, in particular on an Internet forum. I am thinking of high-resolution scanning or macro-reprography of the finished print.
- Compare inkjet (Giclee) prints to C-prints (sometimes called Lambda or LightJet prints) and the good old Cibachromes. What defines the organic look (or the digital look) often alluded to in film versus digital debates.
- Determine the effective resolution of the printer, both with and without external RIP (raster image processor) software. Use test charts for resolution and real-world images to check a loss of tonal separation and micro-contrast. Note that inkjet printers use a dithering process to make up intermediate colors and tones. This requires a much higher dpi resolution of the printer compared to the ppi resolution of the file. On the other hand, imperfectly overlapping ink dots may even help to create the impression of subpixel detail rendering.
- Test the requirements on the paper. It must have a very fine fiber structure to not limit resolution, but also a high enough density to take a large quantity of ink.
- Compare a stitch of 4-6 (possibly not tack sharp) images to an upsized, perfectly sharp image captured at 36 MP**. Study the difference in apparent image quality of the final print depending on the mathematical model used for uprezzing, i.e., BiCubic, Lanchoz, Spline, etc. Is uprezzing best done in Photoshop or in the RAW developer?
- In case of film scanning, compare a scan at the resolution limit of the slide film (about 2400 dpi) to a largely oversampled scan on a high-end drum scanner. Is it worth to spend 150 $ (US) for a high-end drum scan or are the Flextight scans sufficient for a 29” print?
- Define criteria for over sharpening other than the obvious halos around fine detailed structures. Can added noise in Photoshop help to make the print look more organic? Conversely, does (even minimal) noise reduction result in smudging?
Lots of work ahead. SR
** A similar sized print from 8×10″ looks richer than one from 4×5″, which in turn looks better than a print from medium format, although the micro-level acuity goes the other way. But this could mean that a 60+ MP full frame sensor (equivalent of 4 Olympus OM-D Micro-four-thirds) yields better prints than the 36 MP sensor of the D800E in spite of poorer pixel acuity, increased noise, and limiting lens performance. Why is this not produced? I guess, the camera designers feel obliged to satisfy the pixel peepers, and therefore we find this aggressive, in-camera sharpening. As my review of the Leaf Credo has shown, the manufacturers of the medium-format backs, serving the professional market, seem to be less stressed in this regard.